A New Approach to Detect and Classify Graphic Primitives in Engineering Drawings

This paper presents a method to recognize and classify graphics in engineering drawing. The method consists of skeletonizing the input image, followed by segmentation of lines, arcs, and circles. Extracted line segments are classified into object lines, centerlines, dashed lines, and section lines, along with detection of arrowheads attached with lines and arcs. The stages of segmentation and classification are based on novel ideas of dominant chain code in digital geometry, unreliable area, and real-geometric analysis. Experimental results on various datasets demonstrate the efficiency and robustness of the proposed method.